how-possibly explanations

This project investigates the modal features of artificial intelligence (AI) systems. These features are unexplored, yet instantiated in current AI practices. For example, some machine learning models 1) allegedly help to provide how-possibly explanations of phenomena, 2) are trained on synthetic (non-actual) data, or 3) aim to provide feasible explanations.

Scientific models provide modal information. But how models represent that information and under what conditions that information has epistemic value is contentious. There are many unresolved questions. How can models represent ostensible non-actual systems?

In the current debate on the relation between how-possibly explanations (HPEs) and understanding, two seemingly irreconcilable …